Terahertz technology is known as “one of the top ten technologies to change the future world”, which is of great significance to basic scientific research, national economic development and national defense construction, especially in the future 6G communication. Terahertz source is essential to terahertz technology research and the core component of terahertz application system. In recent years, Resonant Tunneling Diode(RTD) terahertz source has attracted extensive attention because of its small volume, light weight, easy integration, room temperature operation and low power consumption, which opens up a new way for the popularization and application of terahertz. Through literature analysis, this paper reviews the development of RTD terahertz oscillators based on InP and GaN from the aspects of device material technology, main processes and device properties. At present, how to prepare high-performance, mature and stable InP and GaN based RTD terahertz oscillators has always been a research direction.

Determining the variety of traditional Chinese medicine is the first step to ensure the quality of traditional Chinese medicine. In order to explore the rapid identification method of Chinese herbal medicine varieties, the classification and identification of six kinds of Chinese herbal medicine varieties are studied by terahertz spectroscopy combined with pattern recognition. Six kinds of commonly used Chinese herbal medicines such as Baifupian, Rhubarb, Dangshen, Tangerine Peel, Ophiopogon Japonicus and Gastrodia Elata are collected. A total of 420 groups of terahertz spectral data are obtained. Support Vector Machine(SVM), Principal Component Analysis(PCA) combined with SVM, Linear Discriminant Analysis(LDA) combined with SVM are employed in the 0.2-1.5 THz band to qualitatively identify six kinds of traditional Chinese medicine. The results show that the LDA-SVM Chinese herbal medicine variety recognition model based on terahertz spectral data combined with linear discriminant analysis and SVM is the best, with the accuracy of 100% and 98.41% for unknown samples. The LDA-SVM model in this paper bears good identification ability, can quickly and accurately identify the varieties of traditional Chinese medicine, and provides another identification means for the quality control of traditional Chinese medicine.

Reflective terahertz time-domain spectroscopy is employed to test the time-domain spectrum of coatings. The critical threshold to determine whether the coatings are waterborne or solvent is obtained by the ratio of the first reflection peak to the second reflection peak. This method can successfully achieve rapid analysis by placing the coatings in a measuring container without complicated sample pretreatment. The critical threshold will not be affected by disposable paper cups or polyethylene plastic containers, different testers or test time. This provides a fast and accurate method to identify waterborne coatings and solvent coatings in the field of environmental protection supervision.

To deal with the problem of motion parameter estimation for a high-speed maneuvering target under low Signal-to-Noise Ratio(SNR) environment, based on Hough Transform(HT) and Modified Discrete Chirp-Fourier Transform(MDCFT), a novel algorithm, i.e., HMDCFT, is proposed in this paper. It first employs the trajectory feature extracted from the image of range profiles by HT to correct the first-order Range Migration(RM) and resolve the Doppler ambiguity. Then, it adopts MDCFT to compensate the Doppler Frequency Migration(DFM) and estimate the target's motion parameters. After that, it coherently integrates the energy of radar returns along the estimated target moving trajectory. The numerical experiments show that the proposed method has superior anti-noise ability and motion parameter estimation performance. Compared with the envelope cross-correlation function method based on the correlation between range profiles, its anti-noise ability is improved by about 17.6 dB. Even the input SNR of radar echoes is as low as -35 dB, this method can accurately estimate the target's motion parameters and fully focus the energy of the target's echoes.

Low Earth Orbit(LEO) broadband constellation satellite communication is an important supplement to the ground 5G wireless communication system, and it always faces the problems of large number of ground visible satellites and high computational complexity of traditional satellite selection algorithms. In order to achieve efficient satellite grouping selection algorithm, based on the principle of Multiple Input Multiple Output(MIMO) system, the large-scale path loss model is taken as the basis, combined with the decreasing satellite selection algorithm, so as to reduce the computational complexity and approach the optimal capacity performance with faster convergence speed effectively. The algorithm is verified by numerical simulation under the typical LEO constellation system configuration, and provides a reference for the future 5G low-orbit satellite constellation communication transmission scheme design.

An S-band high-power narrow-band random signal source based on magnetron is developed by taking the magnetron’s advantages, including high power, high efficiency, small size and low cost. Theoretically, the spectrum of a magnetron’s output signal is closely related to its power supply. Herein, the output spectra of magnetron under different cathode currents and anode voltages are experimentally measured. Then, a microcontroller is utilized to control the power supply to generate random variation of anode voltage and cathode current in a specific range. In this way, the output microwave signal of the magnetron would also change randomly in a certain range, thus generating a high-power, narrow-band random signal. Experimental results show that the center frequency, average output power of the narrowband random signal are 2.46 GHz and 1.4 kW, respectively, with DC to microwave conversion efficiency higher than 70%. The high-power narrow-band random signal source developed in this paper has potential applications in electronic countermeasures, communication jamming and anti-jamming situations.

Special and irregular structures such as electric traction frame, cutting, station, bridge crossing, open cut tunnel and so on in the high-speed railway scene experience different propagation mechanism and multipath effect on radio wave propagation from other scenes, which makes radio wave present unique loss and fading characteristics, with high spatial heterogeneity and frequency dependence. As a result, the empirical model cannot accurately represent the radio wave propagation characteristics of the railway scene, which has become one of the theoretical bottlenecks restricting the accuracy and efficiency of wireless network planning and the high-quality development of the railway mobile communication system. Aiming at this challenging problem, research on high-precision network planning and optimization system of railway mobile communication is carried out based on multi-source heterogeneous data fusion. High performance Ray Tracing(RT) technology is employed to optimize the network simulation of high-speed railway scenes 930 MHz band of railway communication. The simulation results show that 90% of the received signal strength is 13.9 dB higher than that before optimization, which significantly improves the wireless network quality in complex railway scenes.

A cascading injection locked S-band Voltage-Controlled Oscillator(VCO) array is proposed in this paper. It facilitates the VCO array configuration and enhances its scale by using cascading injection-locked method instead of traditional single order injection method. All VCOs are cascaded by coupling network. At the same time, it is easy to increase the stage of cascaded circuits. A reference signal is injected into the first VCO. The second VCO is locked through the signal coupled from the output of the previous VCO. Each reference signal is injected into the VCO through its voltage tuning port. It locks the output frequency and decreases the phase noise. Multiple VCOs are locked by one signal through coupling in between. Two cascading injection-locked VCO arrays are designed, fabricated, and measured. The VCOs output identical frequencies with similar phase noises. The phase noise of each VCO is near to -105 dBc/Hz, when the phase noise of the source is -107.28 dBc/Hz. The injection-locking method is simple and low-cost, which has future applications in phased arrays.

Microwave heating, as a fast, efficient and clean heating method, has been widely used in materials processing. In this work, the heating characteristics of solder paste in microwave electric field and magnetic field are analyzed according to the microwave dissipation mechanism of metal powder, and the heating effect of solder paste circuit in microwave electric field and microwave magnetic field is studied. The experimental results show that both microwave electric field and microwave magnetic field can quickly heat the solder paste, but the high intensity microwave electric field is prone to excite the plasma and burn the substrate. Contrarily, microwave magnetic field selectively heats the solder paste to achieve rapid heating and melting of the solder joint while keeping the substrate at a low temperature. By comparing the microstructure of the solder joints melted by rapid microwave field with those melted by conventional heating, it was found that the solder joints melted by rapid microwave field heating have extremely thin intermetallic compound thickness, which is conducive to improving the strength of the solder joints. This study provides an excellent solution for the further development of solder paste welding of flex circuits.

Focusing on the electromagnetic information security of keyboard, the working principle and signal characteristics of PS/2 keyboard are analyzed, and a detection method based on deep learning is put forward. This method adapts the Convolution Neural Network(CNN) structure to the electromagnetic leakage signal of the keyboard equipment. Combining with the improved Gradient weighted Class Activation Mapping(Grad-CAM) method, the intelligent recognition and location of keyboard electromagnetic information characteristics are realized. Testing on the electromagnetic signals of four keyboards, the classification accuracy can reach 98%. And the classification accuracy can reach 81% in noisy environment. The experimental results indicate that the improved method has better location performance for the electromagnetic information of keyboards.

The calculation of Grnwald-Letnikov's(G-L) weighted coefficient of first-order approximation has an accurate and fast recursive formula, while the calculation of Lubich's weighted coefficient of high-order approximation has a high complexity and a long calculation time. The approximation order of Lubich operator is proved by Fourier transform, and a class of Weighted Shifted Lubich Difference(WSLD) operators of fourth order approximation is proposed based on shifted Lubich operators. Then, the filtering characteristics of WSLD operator are analyzed from the perspective of digital signal processing, and a fractional digital Finite Impulse Response(FIR) differential filter based on WSLD operator is designed and verified by numerical simulation. Compared with Al-Alaoui and Lubich, the results show that the algorithm using WSLD operator to solve the filter coefficient of fractional digital FIR filter is simple and efficient, and can effectively reduce the influence of Gibbs effect compared with other operators.

Group activity recognition is to identify the common activity of individuals in space. It is closely related not only to group state, but also to the individual spatio-temporal features which can not only describe the semantic information in space, but also reflect the dynamic changes of activity. A group activity recognition method is proposed based on attention mechanism and deep spatio-temporal information. Firstly, the advanced ShuffleAttention is introduced into the two-stream feature extraction network to extract individual appearance and motion information. Secondly, the improved non-local network is employed to extract the deep temporal information. Finally, the individual features are input into the graph convolutional neural network to model the interaction and spatial information, recognizing group activity. The accuracies on Collective Activity Dataset(CAD) and Collective Activity Extended Dataset(CAED) reach 93.6% and 97.8% respectively. Compared with Cohesive Cluster Search(CCS) and Actor Relation Graph(ARG) methods, the accuracy of the proposed method on CAD datasets is 1.2% and 2.6% higher, indicating that the proposed method can effectively extract deep spatio-temporal features and improve the accuracy of group activity recognition.

In order to tackle the problem of“premature”in traditional adaptive genetic algorithms,a new adaptive genetic algorithm based on population diversity is proposed. The key to solve the immature convergence problem is to avoid the loss of population diversity before the algorithm finds the optimal solution. In order to adapt to the changes in population diversity in the evolutionary process, the crossover probability and mutation probability formulas including the variance factor and the population entropy factor are designed. According to the population convergence, the crossover probability and mutation probability are adjusted accordingly to maintain the diversity of the population without destroying the good gene model of the population. By testing standard functions and comparing them with existing algorithms, the results show that the improved algorithm not only can ensure the convergence accuracy, but also increase the convergence speed, and effectively overcome the“premature”problems.

To tackle with the problem of poor visual effects caused by low-resolution of medical ultrasound images, a neural network based image Super-Resolution(SR) reconstruction approach is employed to improve the resolution of medical ultrasound images. Based on the Generative Adversarial Network for Super-Resolution(SRGAN), the structure of the network is changed by reducing two input channels and deleting a residual block. A fuzzy dataset is added and the loss function of the network is improved according to the characteristics of medical ultrasound images, such as gray-scale image and single speckle texture, so that the network is adapted to reconstruct the clear edges with 4 times magnification of medical ultrasound images without artifacts. Comparing the results of the improved SRGAN with the original SRGAN, the Peak Signal-to-Noise Ratio(PSNR) and Structural SIMilarity (SSIM) are increased by 1.792 dB and 3.907% respectively; compared with Bicubic interpolation, the PSNR and SSIM are increased by 2.172% dB and 8.732% respectively.

Optoelectronic Oscillator(OEO) can generate low phase noise microwave signals. In an OEO, the Mach-Zehnder Modulator(MZM) can work either at the quadrature-biased point to minimize the fundamental oscillating signal loss or at the Null-Transmission Point(NTP) to obtain the frequency-doubling signal. In the OEO where the MZM is NTP-biased(NTP-biased OEO), the fundamental signal can be obtained by using an electrical frequency divider so as to maintain oscillation in the loop. In this paper, the phase noise performance of the two OEOs(Q-biased OEO and NTP-biased OEO) is analyzed theoretically. According to the theory, it can be seen that the phase noise performance of the NTP-biased OEO is better than that of the Q-biased OEO. With the simulations, the phase noise floor of the NTP-biased OEO is 3 dB better than that of the Q-biased OEO. Besides, the free spectral range will not be affected by the frequency division in the NTP-biased OEO.

Due to the advantages of low ripples in input and output current, and bidirectional energy flowing for the Cuk converter, it shows good application prospects both in new energy generation and DC micro-grid. Based on the analysis of the converter's working principle, the turn-on and turn-off switching flow graphs are established respectively. Then, the switching branch models realized with multipliers to derive the switching flow graph in the whole switching period. Finally, the small signal model of the converter is obtained by imposing disturbance on the switching branches, and the transfer functions of the Cuk converter are calculated with Mason's gain formula. The correctness of the proposed model is proved by simulation results of the AC analysis on the converter, which is implemented with Power Simulation(PSIM) software. The modeling method in this paper has a high reference value for investigations on high order DC/DC converters' modeling.

In order to meet the requirements of miniaturization and lightweight development of Transmitter/Receiver(T/R) module in active phased array radar, a novel vertical interconnection structure is proposed in this paper. Semi-embedded Ball Grid Array(BGA) vertical interconnection structure is designed by using electromagnetic simulation software HFSS from the point of view of process optimization, combined with the characteristics of Low Temperature Co-fired Ceramic(LTCC) substrate. In addition, the influence of semi-embedded structure on vertical interconnection transmission performance is analyzed. The results show that the return loss is higher than 24 dB and the insertion loss is lower than 0.15 dB in X-band and 20 dB, 0.6 dB in Ku-band. The semi-embedded structure can achieve good microwave transmission performance in a wide band between X-Ku band while optimizing the process.